工业水处理 ›› 2025, Vol. 45 ›› Issue (5): 157-165. doi: 10.19965/j.cnki.iwt.2024-0325

• 试验研究 • 上一篇    

基于优化支持向量回归机的气浮单元水质预测模型

陈霖1,2(), 晏欣1, 唐智和1, 冉照宽1, 李斌莲1, 栾辉1(), 陈春茂2   

  1. 1. 中国石油集团安全环保技术研究院有限公司,北京 102200
    2. 中国石油大学(北京)化学工程与 环境学院,石油石化污染物控制与处理国家重点实验室,北京 102249
  • 收稿日期:2024-06-14 出版日期:2025-05-20 发布日期:2025-05-22
  • 通讯作者: 栾辉
  • 作者简介:

    陈霖(1997— ),在读博士,E-mail:

  • 基金资助:
    中国石油天然气股份有限公司十四五前瞻性基础性战略性技术研究课题(2022DJ6904)

Water quality prediction model of air flotation unit based on optimization support vector regression machine

Lin CHEN1,2(), Xin YAN1, Zhihe TANG1, Zhaokuan RAN1, Binlian LI1, Hui LUAN1(), Chunmao CHEN2   

  1. 1. China Petroleum Group Safety and Environmental Protection Technology Research Institute Co. , Ltd. , Beijing 102200, China
    2. State Key Laboratory of Heavy Oil Processing, College of Chemical Engineering and Environment, China University of Petroleum(Beijing), Beijing 102249, China
  • Received:2024-06-14 Online:2025-05-20 Published:2025-05-22
  • Contact: Hui LUAN

摘要:

为解决炼化污水处理系统气浮单元出水水质获取时滞严重的问题,构建了基于支持向量回归机(SVR)的气浮单元水质预测模型,利用皮尔逊相关系数(PCC)、斯皮尔曼相关系数(SCC)以及平均影响值算法(MIV)对模型输入参数进行降维,在此基础上利用交叉验证算法(K-CV)和网格搜索算法(GSA)对模型进行参数优化。结果表明,气浮单元出水COD和进水NH3-N相关性最强,去除冗余变量,将NH3-N作为模型输入可以有效提升模型预测精度。当惩罚因子c趋近于1,核函数参数g趋近于2 000时,模型预测均方误差(MSE)最小(MSE=0.000 67),预测精度最高;优化后SVR模型决定系数(R 2)和相关性系数(r)分别为0.69和0.85,平均绝对百分比误差(MAPE)为0.05,预测精度远高于传统SVR和经典BP-ANN模型。现场验证结果表明该模型能实现对气浮单元出水水质的有效预测,平均百分比误差<5%,预测时间<1 min,极大程度提高了水质数据的时效性。

关键词: 炼化企业, 污水处理系统, 气浮单元, 支持向量回归机, 水质预测模型

Abstract:

To solve the problem of serious delay of effluent water quality data from air flotation units in refinery wastewater treatment system, a water quality prediction model based on support vector regression(SVR) was developed. Pearson correlation coefficients(PCC), Spearman correlation coefficients(SCC), and mean influence value algorithm(MIV) were used to reduce the dimensionality of the input parameters of the model. On this basis, cross validation algorithm(K-CV) and grid search algorithm(GSA) were used to optimize the parameters of SVR prediction model. The results indicated that the effluent COD and influent NH3-N of the air flotation unit had the strongest correlation. Removing redundant variables and using NH3-N as the model input could effectively improve the prediction accuracy of the model. When the penalty factor c approached 1 and the kernel function parameter g approached 2 000, the mean square error(MSE) of the model prediction was the smallest (MSE=0.000 67) and the accuracy was the highest. The determination coefficient(R 2) and correlation coefficient(r) of the SVR model were 0.69 and 0.85, respectively, with the mean absolute percentage error(MAPE) of 0.05. The prediction accuracy was much higher than that of the SVR model and BP neural network model. The on-site verification results showed that the model could effectively predict the effluent water quality of the air flotation unit, with an average percentage error of <5% and prediction time of <1 min, which greatly improved the timeliness of water quality data.

Key words: petrochemical refineries, wastewater treatment system, air flotation unit, support vector regression, water quality prediction model

中图分类号: